Data processing system for equipment vibration monitoring

Abstract

The invention relates to the field of vibration monitoring data processing, in particular to a data processing system for equipment vibration monitoring, which comprises a data acquisition module, a data processing module, a feature extraction module, an edge processing module and a data transmission module, and is characterized in that the data acquisition module is used for acquiring vibration information of equipment and converting the vibration information into digital signals. And the data processing module is used for preprocessing the input digital signal to ensure the data quality. And the feature extraction module is used for performing feature extraction on the processed digital signal and outputting a feature analysis result. And the edge processing module identifies the input feature analysis result through the convolutional neural network, and classifies and judges the state of the equipment. The data processing module and the feature extraction module are integrated in the system, so that the current state of the equipment can be more accurately judged, and whether to carry out early warning on the upper computer or not is determined according to the judgment result.

Description

Technical Field

[0001] The present invention relates to the field of vibration monitoring data processing, and particularly to a data processing system for device vibration monitoring. Background Art

[0002] With the continuous construction of modern Internet of Things, the maintenance of devices and the guarantee of safety performance are particularly important. In order to ensure the safe operation of devices, when various working devices under the Internet of Things perform their tasks, various electronic information monitoring systems are required to monitor the operation status of the devices. For the vibration information of the devices, the traditional monitoring method generally monitors and gives alarms by setting thresholds. When parameters such as the vibration amplitude and frequency of the device exceed the preset values, it is considered a failure, risk, etc. However, the actual situation is more complex, and it is impossible to accurately judge the problem of device vibration only by setting thresholds. Therefore, there is an urgent need to design a data processing system for device vibration monitoring. Summary of the Invention

[0003] In view of the deficiencies of the prior art, the present invention provides the following technical solutions: A data processing system for device vibration monitoring includes: a data acquisition module, a data processing module, a feature extraction module, an edge processing module, and a data transmission module.

[0004] Specifically, the data acquisition module is used to collect the vibration information of the device and convert it into a digital signal. The data processing module is used to preprocess the input digital signal to ensure data quality. The feature extraction module is used to extract features from the processed digital signal and output the feature analysis result. The edge processing module identifies the input feature analysis result through a convolutional neural network and classifies and judges the device status. The data transmission module is used to transmit the device status and related data to the host computer for display and / or early warning notification.

[0005] As an improvement of the above technical solution, the data acquisition module includes: a data induction module, an auxiliary module, and a signal conditioning module; The data induction module is used to collect the sound information of device vibration, the auxiliary module is used to provide temperature compensation for the data induction module to eliminate the temperature drift error, and the signal conditioning module is used to convert the vibration analog signal into a data signal.

[0006] As an improvement of the above technical solution, the data processing module includes: a filtering module, a denoising module, and a resampling module. After the digital signal is input into the data processing module, it first undergoes digital filtering processing by the filtering module, then undergoes wavelet denoising processing by the denoising module, and finally the sampling rate is adjusted by the resampling module to dynamically optimize the data volume.

[0007] As an improvement to the above technical solution, the filtering module eliminates power frequency interference through an adaptive notch filter to remove power frequency noise in the vibration signal. The denoising module performs wavelet transform on the input signal using the wavelet threshold denoising method and applies a hard threshold or a soft threshold for denoising. The resampling module resamples the data using linear interpolation or polynomial interpolation.

[0008] As an improvement to the above technical solution, the feature extraction module includes a time-domain feature extraction module, a spectrum analysis module, an envelope demodulation module, and a time-frequency domain feature extraction module. The time-domain feature extraction module is at least used to obtain the RMS, peak factor, and kurtosis of the signal. The spectrum analysis module converts the digital signal processed by the data processing module from the time domain to the frequency domain through fast Fourier transform to analyze the energy distribution of the signal at different frequencies. The envelope demodulation module analyzes the potential fault frequencies in the digital signal through the envelope demodulation method. The time-frequency domain feature extraction module is used to extract the instantaneous frequency and MFCC features of the signal.

[0009] As an improvement to the above technical solution, the acquisition of the RMS depends on the following formula:

[0010] where, is each data point in the signal, and N is the number of data points.

[0011] The method for obtaining the peak factor depends on the following formula:

[0012] where, represents the signal at a certain moment of the maximum amplitude.

[0013] The method for obtaining the kurtosis depends on the following formula:

[0014] where, is the signal mean, is the standard deviation.

[0015] As an improvement to the above technical solution, when the spectrum analysis module works, it performs the following steps: S01: Decompose the signal into several intrinsic mode functions through the empirical mode decomposition method and extract the instantaneous frequency through Hilbert transform.

[0016] S02: Perform short-time Fourier transform on the input signal to obtain the time-frequency diagram of the signal.

[0017] S03: Calculate the mel-frequency cepstral coefficients of the signal and extract the spectral features of the signal.

[0018] As an improvement to the above technical solution, the empirical mode decomposition method depends on the following formula:

[0019] where, is the original signal, is the th intrinsic mode function, is the residual signal.

[0020] The Hilbert transform for extracting the instantaneous frequency depends on the following formula:

[0021] where, is the Hilbert transform of the th intrinsic mode function, is the instantaneous frequency.

[0022] The calculation of the mel-frequency cepstral coefficients depends on the following formula:

[0023]

[0024] where, is the mel frequency, is the ordinary frequency, represents the spectral value of the signal in the time domain and frequency domain, representing the frequency domain component in time and mel frequency, represents time, reflecting the time domain information of the signal, represents the discrete cosine transform.

[0025] As an improvement to the above technical solution, the edge processing module includes a micro convolutional neural network; The convolutional neural network uses the time-frequency diagram as input data, extracts local features in the time-frequency diagram through the convolutional layer, and the pooling layer performs downsampling operations to reduce the feature dimension and enhance translational invariance. Finally, the fully connected layer integrates high-order features and completes the classification and recognition of vibration modes..

[0026] As an improvement to the above technical solution, the data transmission module dynamically adjusts the compression ratio of data transmission according to the analysis result of the edge processing module, and processes the data using a data checksum and retransmission mechanism during data transmission to avoid data loss during transmission.

[0027] Advantages of the present invention: By integrating a data processing module and a feature extraction module into the system, not only can the data quality of vibration data signals be improved, but also features can be extracted from the vibration data. Combining with the edge processing module can further enhance the vibration data, thereby identifying the vibration condition of the device. Different from the traditional single threshold, it can learn and identify based on past fault situations, so as to more accurately judge the current state of the device and decide whether to give a warning to the host computer according to the judgment result. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic block diagram of the principle of the present invention; Figure 2 is a schematic block diagram of the principle of the data acquisition module of the present invention; Figure 3 is a schematic block diagram of the principle of the data processing module of the present invention; Figure 4 is a schematic block diagram of the principle of the feature extraction module of the present invention; Figure 5 is a flowchart when the spectrum analysis module of the present invention is working. DETAILED DESCRIPTION OF THE INVENTION

[0029] The following specific examples are used to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0030] In the security monitoring of the Internet of Things system, generally, threshold settings are used for monitoring and alarming. When parameters such as the vibration amplitude and frequency of the device exceed the preset values, it is considered a fault, risk, etc. However, the actual situation is more complex, and it is impossible to accurately judge the problem of device vibration only by threshold settings.

[0031] To solve the above problems, please refer to Figures 1 to 5 , and a data processing system for device vibration monitoring is provided, including: a data acquisition module, a data processing module, a feature extraction module, an edge processing module, and a data transmission module.

[0032] Specifically, the data acquisition module is used to collect the vibration information of the device and convert it into a digital signal. The data processing module is used to preprocess the input digital signal to ensure data quality. The feature extraction module is used to extract features from the processed digital signal and output the feature analysis result. The edge processing module identifies the input feature analysis result through a convolutional neural network and classifies to determine the device status. The data transmission module is used to transmit the device status and related data to the host computer for display and / or warning notification.

[0033] The daily working conditions of the device are collected by the data acquisition module. When the device vibrates, the vibration information of the device is obtained. Usually, the initially obtained vibration information is the analog signal of the sensor. To enable computer recognition, the data acquisition module is also required to convert the data into a digital signal. The converted digital signal is input into the data processing module for preprocessing to improve the data quality of subsequent feature extraction and data analysis through the preprocessing of the data. The feature extraction module extracts features from the processed vibration signal, extracts important features helpful for diagnosing the device status, provides these important features to the edge processing module for analysis and processing, determines the current working state of the device, and transmits the state and related data to the host computer for display. If there is danger, a warning notification is executed; if there is no danger, no warning notification is executed.

[0034] In one embodiment, please refer to Figure 2 , and the data acquisition module is further improved. Specifically, the data acquisition module includes: a data sensing module, an auxiliary module, and a signal conditioning module.

[0035] Among them, the data sensing module is used to collect the sound information of the device vibration, the auxiliary module is used to provide temperature compensation for the data sensing module to eliminate the temperature drift error, and the signal conditioning module is used to convert the vibration analog signal into a data signal.

[0036] The data sensing module usually uses various sensors to detect the device. For the detection of vibration information, a triaxial acceleration sensor can be used, which can more accurately detect the vibration of the device. The temperature drift error is eliminated through the temperature compensation circuit in the auxiliary module to ensure the stability of the sensor at different temperatures. The signal conditioning module usually includes three groups of modules. The first group is the preamplifier circuit, and usually an instrumentation amplifier can be used to amplify the weak signal output by the sensor. The second group is the low-pass filter, which is used to remove the high-frequency noise in the amplified signal to prevent aliasing. The third group is the ADC conversion module, which converts the processed analog signal into a digital signal through the ADC controller. Usually, an ADC controller with a high signal-to-noise ratio is selected.

[0037] After the data collection is completed, these data are transmitted to the data processing module for processing. Specifically, please refer to Figure 3 , the data processing module includes: a filtering module, a denoising module, and a resampling module. After the digital signal is input into the data processing module, it first undergoes digital filtering processing by the filtering module, then undergoes wavelet denoising processing by the denoising module, and finally the sampling rate is adjusted by the resampling module to dynamically optimize the data volume.

[0038] Among them, the filtering module eliminates power frequency interference through an adaptive notch filter, which can adaptively adjust its parameters according to the signal spectrum to remove power frequency noise in the vibration signal. The denoising module uses the wavelet threshold denoising method to perform wavelet transform on the input signal and applies a hard threshold or a soft threshold for denoising. The resampling module uses linear interpolation or polynomial interpolation to resample the data.

[0039] After the input vibration signal is processed by the filtering module and the denoising module, the effective vibration characteristics are retained. After resampling, the signal with the sampling rate optimized can reduce the data volume without losing important information, thereby reducing the complexity of subsequent processing.

[0040] After the data processing is completed, it is necessary to extract the key features in the data. Specifically, please refer to Figure 4 , the feature extraction module includes a time-domain feature extraction module, a spectrum analysis module, an envelope demodulation module, and a time-frequency domain feature extraction module. The time-domain feature extraction module is at least used to obtain the RMS, peak factor, and kurtosis of the signal. The spectrum analysis module converts the digital signal processed by the data processing module from the time domain to the frequency domain through a fast Fourier transform to analyze the energy distribution of the signal at different frequencies. The envelope demodulation module analyzes the potential fault frequencies in the digital signal through the envelope demodulation method. The time-frequency domain feature extraction module is used to extract the instantaneous frequency and MFCC features of the signal.

[0041] First, extract the information of the time-domain features. For the time-domain features, usually the RMS (root mean square), peak factor, and kurtosis are extracted. Among them, the RMS is used to measure the overall energy of the signal, the peak factor is used to measure the sharpness of the signal, and the kurtosis is used to measure the peak distribution of the signal. The extraction of these parameters usually depends on specific calculations, as follows: The acquisition of the RMS depends on the following formula:

[0042] Among them, is each data point in the signal, and N is the number of data points; The acquisition method of the peak factor depends on the following formula:

[0043] Among them, represents the maximum amplitude of the signal at a certain moment.

[0044] The method for obtaining the kurtosis depends on the following formula:

[0045] Among them, is the signal mean value, is the standard deviation.

[0046] By extracting the RMS, peak factor, and kurtosis, the vibration amplitude and distribution characteristics of the vibration signal can be described, and these characteristics can play a key role in subsequent analyses. After the extraction of time-domain features is completed, the frequency-domain features also need to be processed. Specifically, please refer to Figure 5 , and when the spectrum analysis module works, it performs the following steps: S01: Decompose the signal into several intrinsic mode functions by the empirical mode decomposition method, and extract the instantaneous frequency through the Hilbert transform.

[0047] The Hilbert transform can be used to process non-stationary signals. Therefore, for non-linear and non-stationary vibration signals, the use of the Hilbert transform can achieve better processing effects. The addition of the empirical mode decomposition can adaptively extract the vibration modes of different frequencies in the signal through the decomposition of local time scales (such as the extraction and decomposition of high-frequency and low-frequency vibrations). Then, by performing the Hilbert transform on each intrinsic mode function to generate the corresponding analytic signal, the instantaneous frequency and amplitude of each intrinsic mode function can be integrated into the time-frequency energy distribution, intuitively displaying the dynamic changes of the signal energy in time and frequency. For the calculation of the corresponding parameters of the empirical mode decomposition method and the Hilbert transform, it depends on the following formulas: The empirical mode decomposition method depends on the following formula:

[0048] Among them, is the original signal, is the th intrinsic mode function, is the residual signal; The Hilbert transform for extracting the instantaneous amplitude depends on the following formula

[0049] Among them, is the Hilbert transform of the th intrinsic mode function, is the instantaneous amplitude.

[0050] The Hilbert transform for extracting the instantaneous frequency depends on the following formula:

[0051] where is the Hilbert transform of the th intrinsic mode function, and is the instantaneous frequency.

[0052] To more intuitively display the data, step S02 is performed as follows: S02: Perform a short-time Fourier transform on the input signal to obtain the time-frequency diagram of the signal.

[0053] The signal is segmented by a sliding window, and the Fourier transform is performed on each segment of the signal to generate a two-dimensional time-frequency diagram (time × frequency), intuitively showing the distribution of the signal energy over time and frequency, and providing a single-pin spectral input for the subsequent calculation of cepstral coefficients.

[0054] S03: Calculate the mel-frequency cepstral coefficients of the signal to extract the spectral features of the signal.

[0055] Specifically, the calculation of the mel-frequency cepstral coefficients depends on the following formula:

[0056]

[0057] where is the mel frequency, is the ordinary frequency, represents the spectral value of the signal in the time domain and frequency domain, representing the frequency domain component in time and mel frequency, represents time, reflecting the time domain information of the signal, represents the discrete cosine transform.

[0058] To classify the features extracted above, specifically, the edge processing module includes a micro convolutional neural network. The convolutional neural network uses the time-frequency diagram as the input data, extracts local features in the time-frequency diagram through the convolutional layer, and the pooling layer performs downsampling operations to reduce the feature dimension and enhance translational invariance. Finally, the fully connected layer integrates high-order features and completes the classification and recognition of vibration modes.

[0059] Among them, the micro convolutional neural network is quantized through the TensorFlowLite tool, which can optimize the inference speed and memory occupancy with fewer parameters. During model training, it can be trained for the vibration data of specific devices, such as bearings. By using the vibration signals under different conditions such as normal operation, imbalance, looseness, and damage of the bearings as training data for training, a micro convolutional neural network for bearing working state recognition can be obtained.

[0060] When the time-frequency map is input into the micro convolutional neural network, normalization, data augmentation and other processing means are usually performed on the time-frequency map to ensure the accuracy of the input data, so as to realize the accurate judgment of the working state of the device. When the model inference result is abnormal, an alarm and notification will be sent.

[0061] In addition, during the data transmission process, the data transmission module dynamically adjusts the compression ratio of data transmission according to the analysis result of the edge processing module, and adopts a data checksum and retransmission mechanism to process the data during the data transmission process to avoid data loss during the transmission process.

[0062] Usually, the MQTT or CoAP protocol is adopted at the protocol level to ensure low latency and high reliability, and wireless communication methods such as Wi-Fi, LoRa or 5G are usually adopted during transmission. The data will not only be sent to the host computer for display, but also be synchronously transmitted to the cloud. If there is an abnormality in the sent signal, an alarm signal will also be sent.

[0063] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A data processing system for device vibration monitoring, characterized in that, Including: A data acquisition module, which is used to collect the vibration information of the device and convert it into a digital signal; A data processing module, which is used to preprocess the input digital signal to ensure data quality; A feature extraction module, which is used to extract features from the processed digital signal and output a feature analysis result; An edge processing module, which identifies the input feature analysis result through a convolutional neural network and classifies and judges the device state; A data transmission module, which is used to transmit the device state and related data to the host computer for display and / or warning notification.

2. The data processing system for device vibration monitoring according to claim 1, wherein: The data acquisition module includes: a data sensing module, an auxiliary module, and a signal conditioning module; The data sensing module is used to collect the sound information of the device vibration, the auxiliary module is used to provide temperature compensation for the data sensing module to eliminate the temperature drift error, and the signal conditioning module is used to convert the vibration analog signal into a data signal.

3. A data processing system for device vibration monitoring according to claim 1, characterized in that: The data processing module includes: a filtering module, a denoising module, and a resampling module; After the digital signal is input into the data processing module, it first undergoes digital filtering processing by the filtering module, then undergoes wavelet denoising processing by the denoising module, and finally adjusts the sampling rate through the resampling module to dynamically optimize the data volume.

4. A data processing system for device vibration monitoring according to claim 3, characterized in that: The filtering module eliminates power frequency interference through an adaptive notch filter and removes power frequency noise in the vibration signal. The denoising module performs wavelet transform on the input signal using the wavelet threshold denoising method and applies a hard threshold or a soft threshold for denoising. The resampling module resamples the data using linear interpolation or polynomial interpolation.

5. The data processing system for device vibration monitoring according to claim 1, characterized in that: The feature extraction module includes a time-domain feature extraction module, a spectrum analysis module, an envelope demodulation module, and a time-frequency domain feature extraction module; The time-domain feature extraction module is at least used to obtain the RMS, peak factor, and kurtosis of the signal. The spectrum analysis module converts the digital signal processed by the data processing module from the time domain to the frequency domain through a fast Fourier transform and analyzes the energy distribution of the signal at different frequencies. The envelope demodulation module analyzes the potential fault frequencies in the digital signal through the envelope demodulation method. The time-frequency domain feature extraction module is used to extract the instantaneous frequency and MFCC features of the signal.

6. The data processing system for device vibration monitoring according to claim 5, wherein: The acquisition of the RMS depends on the following formula: where x i is each data point in the signal, and N is the number of data points; The acquisition method of the peak factor depends on the following formula: Where, max(|x(t)|) represents the maximum amplitude of the signal x(t) at a certain moment; The acquisition method of the kurtosis depends on the following formula: Where, δ is the signal mean, and ε is the standard deviation.

7. A data processing system for device vibration monitoring according to claim 5, characterized in that: When the spectrum analysis module works, it performs the following steps: S01: Decompose the signal into several intrinsic mode functions through the empirical mode decomposition method and extract the instantaneous frequency through the Hilbert transform; S02: Perform a short-time Fourier transform on the input signal to obtain the time-frequency diagram of the signal; S03: Calculate the mel-frequency cepstral coefficients of the signal and extract the spectral features of the signal.

8. A data processing system for device vibration monitoring according to claim 7, characterized in that: The empirical mode decomposition method depends on the following formula: where x(t) is the original signal, and IMF k (t) is the k-th intrinsic mode function, and r M (t) is the residual signal; The extraction of the instantaneous frequency by the Hilbert transform depends on the following formula: where H{IMF k (t)} is the Hilbert transform of the k-th intrinsic mode function, and f(t) is the instantaneous frequency; The calculation of the mel-frequency cepstral coefficients depends on the following formula: MFCC = DCT(log(|X(t,f m |))) Among them, f m is the Mel frequency, f is the ordinary frequency, X(t, f m |) represents the spectral value of the signal in the time domain and the frequency domain, represents the frequency domain component at time and Mel frequency, t represents time, reflecting the time domain information of the signal, and DCT represents the discrete cosine transform.

9. The data processing system for device vibration monitoring according to claim 1, characterized in that: The edge processing module includes a micro convolutional neural network; The convolutional neural network uses the time-frequency diagram as the input data, extracts local features in the time-frequency diagram through the convolutional layer, and the pooling layer performs downsampling operations to reduce the feature dimension and enhance translational invariance. Finally, the fully connected layer integrates high-order features and completes the classification and recognition of vibration modes.

10. A data processing system for device vibration monitoring according to claim 1, characterized in that: The data transmission module dynamically adjusts the compression ratio of data transmission according to the analysis results of the edge processing module, and uses a data checksum and retransmission mechanism to process the data during data transmission to avoid data loss during transmission.

Images